Learning Language Models of Movie Characters

We investigate learning language models of individual movie characters. We train a recurrent neural net based model on a large dataset of movie scripts with no character specificity to learn a general dialogue model first. Then, we transfer the parameters from this pretrained model to initialize another model and learn a character specific model from a single show. We measure the performance by using perplexity on the general model and relative change in perplexity by switching out character models for the character specific model. Our results suggest that it is possible to learn and evaluate character specific language models.